1 /*

        2  * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.

        3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.

        4  *

        5  * This code is free software; you can redistribute it and/or modify it

        6  * under the terms of the GNU General Public License version 2 only, as

        7  * published by the Free Software Foundation.

        8  *

        9  * This code is distributed in the hope that it will be useful, but WITHOUT

       10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or

       11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License

       12  * version 2 for more details (a copy is included in the LICENSE file that

       13  * accompanied this code).

       14  *

       15  * You should have received a copy of the GNU General Public License version

       16  * 2 along with this work; if not, write to the Free Software Foundation,

       17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.

       18  *

       19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA

       20  * or visit www.oracle.com if you need additional information or have any

       21  * questions.

       22  *

       23  */

       24 

       25 // no precompiled headers

       26 #include "assembler_x86.inline.hpp"

       27 #include "classfile/classLoader.hpp"

       28 #include "classfile/systemDictionary.hpp"

       29 #include "classfile/vmSymbols.hpp"

       30 #include "code/icBuffer.hpp"

       31 #include "code/vtableStubs.hpp"

       32 #include "interpreter/interpreter.hpp"

       33 #include "jvm_linux.h"

       34 #include "memory/allocation.inline.hpp"

       35 #include "mutex_linux.inline.hpp"

       36 #include "nativeInst_x86.hpp"

       37 #include "os_share_linux.hpp"

       38 #include "prims/jniFastGetField.hpp"

       39 #include "prims/jvm.h"

       40 #include "prims/jvm_misc.hpp"

       41 #include "runtime/arguments.hpp"

       42 #include "runtime/extendedPC.hpp"

       43 #include "runtime/frame.inline.hpp"

       44 #include "runtime/interfaceSupport.hpp"

       45 #include "runtime/java.hpp"

       46 #include "runtime/javaCalls.hpp"

       47 #include "runtime/mutexLocker.hpp"

       48 #include "runtime/osThread.hpp"

       49 #include "runtime/sharedRuntime.hpp"

       50 #include "runtime/stubRoutines.hpp"

       51 #include "runtime/timer.hpp"

       52 #include "thread_linux.inline.hpp"

       53 #include "utilities/events.hpp"

       54 #include "utilities/vmError.hpp"

       55 #ifdef COMPILER1

       56 #include "c1/c1_Runtime1.hpp"

       57 #endif

       58 #ifdef COMPILER2

       59 #include "opto/runtime.hpp"

       60 #endif

       61 

       62 // put OS-includes here

       63 # include <sys/types.h>

       64 # include <sys/mman.h>

       65 # include <pthread.h>

       66 # include <signal.h>

       67 # include <errno.h>

       68 # include <dlfcn.h>

       69 # include <stdlib.h>

       70 # include <stdio.h>

       71 # include <unistd.h>

       72 # include <sys/resource.h>

       73 # include <pthread.h>

       74 # include <sys/stat.h>

       75 # include <sys/time.h>

       76 # include <sys/utsname.h>

       77 # include <sys/socket.h>

       78 # include <sys/wait.h>

       79 # include <pwd.h>

       80 # include <poll.h>

       81 # include <ucontext.h>

       82 # include <fpu_control.h>

       83 

       84 #ifdef AMD64

       85 #define REG_SP REG_RSP

       86 #define REG_PC REG_RIP

       87 #define REG_FP REG_RBP

       88 #define SPELL_REG_SP "rsp"

       89 #define SPELL_REG_FP "rbp"

       90 #else

       91 #define REG_SP REG_UESP

       92 #define REG_PC REG_EIP

       93 #define REG_FP REG_EBP

       94 #define SPELL_REG_SP "esp"

       95 #define SPELL_REG_FP "ebp"

       96 #endif // AMD64

       97 

       98 address os::current_stack_pointer() {

       99 #ifdef SPARC_WORKS

      100   register void *esp;

      101   __asm__("mov %%"SPELL_REG_SP", %0":"=r"(esp));

      102   return (address) ((char*)esp + sizeof(long)*2);

      103 #else

      104   register void *esp __asm__ (SPELL_REG_SP);

      105   return (address) esp;

      106 #endif

      107 }

      108 

      109 char* os::non_memory_address_word() {

      110   // Must never look like an address returned by reserve_memory,

      111   // even in its subfields (as defined by the CPU immediate fields,

      112   // if the CPU splits constants across multiple instructions).

      113 

      114   return (char*) -1;

      115 }

      116 

      117 void os::initialize_thread() {

      118 // Nothing to do.

      119 }

      120 

      121 address os::Linux::ucontext_get_pc(ucontext_t * uc) {

      122   return (address)uc->uc_mcontext.gregs[REG_PC];

      123 }

      124 

      125 intptr_t* os::Linux::ucontext_get_sp(ucontext_t * uc) {

      126   return (intptr_t*)uc->uc_mcontext.gregs[REG_SP];

      127 }

      128 

      129 intptr_t* os::Linux::ucontext_get_fp(ucontext_t * uc) {

      130   return (intptr_t*)uc->uc_mcontext.gregs[REG_FP];

      131 }

      132 

      133 // For Forte Analyzer AsyncGetCallTrace profiling support - thread

      134 // is currently interrupted by SIGPROF.

      135 // os::Solaris::fetch_frame_from_ucontext() tries to skip nested signal

      136 // frames. Currently we don't do that on Linux, so it's the same as

      137 // os::fetch_frame_from_context().

      138 ExtendedPC os::Linux::fetch_frame_from_ucontext(Thread* thread,

      139   ucontext_t* uc, intptr_t** ret_sp, intptr_t** ret_fp) {

      140 

      141   assert(thread != NULL, "just checking");

      142   assert(ret_sp != NULL, "just checking");

      143   assert(ret_fp != NULL, "just checking");

      144 

      145   return os::fetch_frame_from_context(uc, ret_sp, ret_fp);

      146 }

      147 

      148 ExtendedPC os::fetch_frame_from_context(void* ucVoid,

      149                     intptr_t** ret_sp, intptr_t** ret_fp) {

      150 

      151   ExtendedPC  epc;

      152   ucontext_t* uc = (ucontext_t*)ucVoid;

      153 

      154   if (uc != NULL) {

      155     epc = ExtendedPC(os::Linux::ucontext_get_pc(uc));

      156     if (ret_sp) *ret_sp = os::Linux::ucontext_get_sp(uc);

      157     if (ret_fp) *ret_fp = os::Linux::ucontext_get_fp(uc);

      158   } else {

      159     // construct empty ExtendedPC for return value checking

      160     epc = ExtendedPC(NULL);

      161     if (ret_sp) *ret_sp = (intptr_t *)NULL;

      162     if (ret_fp) *ret_fp = (intptr_t *)NULL;

      163   }

      164 

      165   return epc;

      166 }

      167 

      168 frame os::fetch_frame_from_context(void* ucVoid) {

      169   intptr_t* sp;

      170   intptr_t* fp;

      171   ExtendedPC epc = fetch_frame_from_context(ucVoid, &sp, &fp);

      172   return frame(sp, fp, epc.pc());

      173 }

      174 

      175 // By default, gcc always save frame pointer (%ebp/%rbp) on stack. It may get

      176 // turned off by -fomit-frame-pointer,

      177 frame os::get_sender_for_C_frame(frame* fr) {

      178   return frame(fr->sender_sp(), fr->link(), fr->sender_pc());

      179 }

      180 

      181 intptr_t* _get_previous_fp() {

      182 #ifdef SPARC_WORKS

      183   register intptr_t **ebp;

      184   __asm__("mov %%"SPELL_REG_FP", %0":"=r"(ebp));

      185 #else

      186   register intptr_t **ebp __asm__ (SPELL_REG_FP);

      187 #endif

      188   return (intptr_t*) *ebp;   // we want what it points to.

      189 }

      190 

      191 

      192 frame os::current_frame() {

      193   intptr_t* fp = _get_previous_fp();

      194   frame myframe((intptr_t*)os::current_stack_pointer(),

      195                 (intptr_t*)fp,

      196                 CAST_FROM_FN_PTR(address, os::current_frame));

      197   if (os::is_first_C_frame(&myframe)) {

      198     // stack is not walkable

      199     return frame(NULL, NULL, NULL);

      200   } else {

      201     return os::get_sender_for_C_frame(&myframe);

      202   }

      203 }

      204 

      205 // Utility functions

      206 

      207 // From IA32 System Programming Guide

      208 enum {

      209   trap_page_fault = 0xE

      210 };

      211 

      212 extern "C" void Fetch32PFI () ;

      213 extern "C" void Fetch32Resume () ;

      214 #ifdef AMD64

      215 extern "C" void FetchNPFI () ;

      216 extern "C" void FetchNResume () ;

      217 #endif // AMD64

      218 

      219 extern "C" JNIEXPORT int

      220 JVM_handle_linux_signal(int sig,

      221                         siginfo_t* info,

      222                         void* ucVoid,

      223                         int abort_if_unrecognized) {

      224   ucontext_t* uc = (ucontext_t*) ucVoid;

      225 

      226   Thread* t = ThreadLocalStorage::get_thread_slow();

      227 

      228   SignalHandlerMark shm(t);

      229 

      230   // Note: it's not uncommon that JNI code uses signal/sigset to install

      231   // then restore certain signal handler (e.g. to temporarily block SIGPIPE,

      232   // or have a SIGILL handler when detecting CPU type). When that happens,

      233   // JVM_handle_linux_signal() might be invoked with junk info/ucVoid. To

      234   // avoid unnecessary crash when libjsig is not preloaded, try handle signals

      235   // that do not require siginfo/ucontext first.

      236 

      237   if (sig == SIGPIPE || sig == SIGXFSZ) {

      238     // allow chained handler to go first

      239     if (os::Linux::chained_handler(sig, info, ucVoid)) {

      240       return true;

      241     } else {

      242       if (PrintMiscellaneous && (WizardMode || Verbose)) {

      243         char buf[64];

      244         warning("Ignoring %s - see bugs 4229104 or 646499219",

      245                 os::exception_name(sig, buf, sizeof(buf)));

      246       }

      247       return true;

      248     }

      249   }

      250 

      251   JavaThread* thread = NULL;

      252   VMThread* vmthread = NULL;

      253   if (os::Linux::signal_handlers_are_installed) {

      254     if (t != NULL ){

      255       if(t->is_Java_thread()) {

      256         thread = (JavaThread*)t;

      257       }

      258       else if(t->is_VM_thread()){

      259         vmthread = (VMThread *)t;

      260       }

      261     }

      262   }

      263 /*

      264   NOTE: does not seem to work on linux.

      265   if (info == NULL || info->si_code <= 0 || info->si_code == SI_NOINFO) {

      266     // can't decode this kind of signal

      267     info = NULL;

      268   } else {

      269     assert(sig == info->si_signo, "bad siginfo");

      270   }

      271 */

      272   // decide if this trap can be handled by a stub

      273   address stub = NULL;

      274 

      275   address pc          = NULL;

      276 

      277   //%note os_trap_1

      278   if (info != NULL && uc != NULL && thread != NULL) {

      279     pc = (address) os::Linux::ucontext_get_pc(uc);

      280 

      281     if (pc == (address) Fetch32PFI) {

      282        uc->uc_mcontext.gregs[REG_PC] = intptr_t(Fetch32Resume) ;

      283        return 1 ;

      284     }

      285 #ifdef AMD64

      286     if (pc == (address) FetchNPFI) {

      287        uc->uc_mcontext.gregs[REG_PC] = intptr_t (FetchNResume) ;

      288        return 1 ;

      289     }

      290 #endif // AMD64

      291 

      292     // Handle ALL stack overflow variations here

      293     if (sig == SIGSEGV) {

      294       address addr = (address) info->si_addr;

      295 

      296       // check if fault address is within thread stack

      297       if (addr < thread->stack_base() &&

      298           addr >= thread->stack_base() - thread->stack_size()) {

      299         // stack overflow

      300         if (thread->in_stack_yellow_zone(addr)) {

      301           thread->disable_stack_yellow_zone();

      302           if (thread->thread_state() == _thread_in_Java) {

      303             // Throw a stack overflow exception.  Guard pages will be reenabled

      304             // while unwinding the stack.

      305             stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::STACK_OVERFLOW);

      306           } else {

      307             // Thread was in the vm or native code.  Return and try to finish.

      308             return 1;

      309           }

      310         } else if (thread->in_stack_red_zone(addr)) {

      311           // Fatal red zone violation.  Disable the guard pages and fall through

      312           // to handle_unexpected_exception way down below.

      313           thread->disable_stack_red_zone();

      314           tty->print_raw_cr("An irrecoverable stack overflow has occurred.");

      315         } else {

      316           // Accessing stack address below sp may cause SEGV if current

      317           // thread has MAP_GROWSDOWN stack. This should only happen when

      318           // current thread was created by user code with MAP_GROWSDOWN flag

      319           // and then attached to VM. See notes in os_linux.cpp.

      320           if (thread->osthread()->expanding_stack() == 0) {

      321              thread->osthread()->set_expanding_stack();

      322              if (os::Linux::manually_expand_stack(thread, addr)) {

      323                thread->osthread()->clear_expanding_stack();

      324                return 1;

      325              }

      326              thread->osthread()->clear_expanding_stack();

      327           } else {

      328              fatal("recursive segv. expanding stack.");

      329           }

      330         }

      331       }

      332     }

      333 

      334     if (thread->thread_state() == _thread_in_Java) {

      335       // Java thread running in Java code => find exception handler if any

      336       // a fault inside compiled code, the interpreter, or a stub

      337 

      338       if (sig == SIGSEGV && os::is_poll_address((address)info->si_addr)) {

      339         stub = SharedRuntime::get_poll_stub(pc);

      340       } else if (sig == SIGBUS /* && info->si_code == BUS_OBJERR */) {

      341         // BugId 4454115: A read from a MappedByteBuffer can fault

      342         // here if the underlying file has been truncated.

      343         // Do not crash the VM in such a case.

      344         CodeBlob* cb = CodeCache::find_blob_unsafe(pc);

      345         nmethod* nm = cb->is_nmethod() ? (nmethod*)cb : NULL;

      346         if (nm != NULL && nm->has_unsafe_access()) {

      347           stub = StubRoutines::handler_for_unsafe_access();

      348         }

      349       }

      350       else

      351 

      352 #ifdef AMD64

      353       if (sig == SIGFPE  &&

      354           (info->si_code == FPE_INTDIV || info->si_code == FPE_FLTDIV)) {

      355         stub =

      356           SharedRuntime::

      357           continuation_for_implicit_exception(thread,

      358                                               pc,

      359                                               SharedRuntime::

      360                                               IMPLICIT_DIVIDE_BY_ZERO);

      361 #else

      362       if (sig == SIGFPE /* && info->si_code == FPE_INTDIV */) {

      363         // HACK: si_code does not work on linux 2.2.12-20!!!

      364         int op = pc[0];

      365         if (op == 0xDB) {

      366           // FIST

      367           // TODO: The encoding of D2I in i486.ad can cause an exception

      368           // prior to the fist instruction if there was an invalid operation

      369           // pending. We want to dismiss that exception. From the win_32

      370           // side it also seems that if it really was the fist causing

      371           // the exception that we do the d2i by hand with different

      372           // rounding. Seems kind of weird.

      373           // NOTE: that we take the exception at the NEXT floating point instruction.

      374           assert(pc[0] == 0xDB, "not a FIST opcode");

      375           assert(pc[1] == 0x14, "not a FIST opcode");

      376           assert(pc[2] == 0x24, "not a FIST opcode");

      377           return true;

      378         } else if (op == 0xF7) {

      379           // IDIV

      380           stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::IMPLICIT_DIVIDE_BY_ZERO);

      381         } else {

      382           // TODO: handle more cases if we are using other x86 instructions

      383           //   that can generate SIGFPE signal on linux.

      384           tty->print_cr("unknown opcode 0x%X with SIGFPE.", op);

      385           fatal("please update this code.");

      386         }

      387 #endif // AMD64

      388       } else if (sig == SIGSEGV &&

      389                !MacroAssembler::needs_explicit_null_check((intptr_t)info->si_addr)) {

      390           // Determination of interpreter/vtable stub/compiled code null exception

      391           stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::IMPLICIT_NULL);

      392       }

      393     } else if (thread->thread_state() == _thread_in_vm &&

      394                sig == SIGBUS && /* info->si_code == BUS_OBJERR && */

      395                thread->doing_unsafe_access()) {

      396         stub = StubRoutines::handler_for_unsafe_access();

      397     }

      398 

      399     // jni_fast_Get<Primitive>Field can trap at certain pc's if a GC kicks in

      400     // and the heap gets shrunk before the field access.

      401     if ((sig == SIGSEGV) || (sig == SIGBUS)) {

      402       address addr = JNI_FastGetField::find_slowcase_pc(pc);

      403       if (addr != (address)-1) {

      404         stub = addr;

      405       }

      406     }

      407 

      408     // Check to see if we caught the safepoint code in the

      409     // process of write protecting the memory serialization page.

      410     // It write enables the page immediately after protecting it

      411     // so we can just return to retry the write.

      412     if ((sig == SIGSEGV) &&

      413         os::is_memory_serialize_page(thread, (address) info->si_addr)) {

      414       // Block current thread until the memory serialize page permission restored.

      415       os::block_on_serialize_page_trap();

      416       return true;

      417     }

      418   }

      419 

      420 #ifndef AMD64

      421   // Execution protection violation

      422   //

      423   // This should be kept as the last step in the triage.  We don't

      424   // have a dedicated trap number for a no-execute fault, so be

      425   // conservative and allow other handlers the first shot.

      426   //

      427   // Note: We don't test that info->si_code == SEGV_ACCERR here.

      428   // this si_code is so generic that it is almost meaningless; and

      429   // the si_code for this condition may change in the future.

      430   // Furthermore, a false-positive should be harmless.

      431   if (UnguardOnExecutionViolation > 0 &&

      432       (sig == SIGSEGV || sig == SIGBUS) &&

      433       uc->uc_mcontext.gregs[REG_TRAPNO] == trap_page_fault) {

      434     int page_size = os::vm_page_size();

      435     address addr = (address) info->si_addr;

      436     address pc = os::Linux::ucontext_get_pc(uc);

      437     // Make sure the pc and the faulting address are sane.

      438     //

      439     // If an instruction spans a page boundary, and the page containing

      440     // the beginning of the instruction is executable but the following

      441     // page is not, the pc and the faulting address might be slightly

      442     // different - we still want to unguard the 2nd page in this case.

      443     //

      444     // 15 bytes seems to be a (very) safe value for max instruction size.

      445     bool pc_is_near_addr =

      446       (pointer_delta((void*) addr, (void*) pc, sizeof(char)) < 15);

      447     bool instr_spans_page_boundary =

      448       (align_size_down((intptr_t) pc ^ (intptr_t) addr,

      449                        (intptr_t) page_size) > 0);

      450 

      451     if (pc == addr || (pc_is_near_addr && instr_spans_page_boundary)) {

      452       static volatile address last_addr =

      453         (address) os::non_memory_address_word();

      454 

      455       // In conservative mode, don't unguard unless the address is in the VM

      456       if (addr != last_addr &&

      457           (UnguardOnExecutionViolation > 1 || os::address_is_in_vm(addr))) {

      458 

      459         // Set memory to RWX and retry

      460         address page_start =

      461           (address) align_size_down((intptr_t) addr, (intptr_t) page_size);

      462         bool res = os::protect_memory((char*) page_start, page_size,

      463                                       os::MEM_PROT_RWX);

      464 

      465         if (PrintMiscellaneous && Verbose) {

      466           char buf[256];

      467           jio_snprintf(buf, sizeof(buf), "Execution protection violation "

      468                        "at " INTPTR_FORMAT

      469                        ", unguarding " INTPTR_FORMAT ": %s, errno=%d", addr,

      470                        page_start, (res ? "success" : "failed"), errno);

      471           tty->print_raw_cr(buf);

      472         }

      473         stub = pc;

      474 

      475         // Set last_addr so if we fault again at the same address, we don't end

      476         // up in an endless loop.

      477         //

      478         // There are two potential complications here.  Two threads trapping at

      479         // the same address at the same time could cause one of the threads to

      480         // think it already unguarded, and abort the VM.  Likely very rare.

      481         //

      482         // The other race involves two threads alternately trapping at

      483         // different addresses and failing to unguard the page, resulting in

      484         // an endless loop.  This condition is probably even more unlikely than

      485         // the first.

      486         //

      487         // Although both cases could be avoided by using locks or thread local

      488         // last_addr, these solutions are unnecessary complication: this

      489         // handler is a best-effort safety net, not a complete solution.  It is

      490         // disabled by default and should only be used as a workaround in case

      491         // we missed any no-execute-unsafe VM code.

      492 

      493         last_addr = addr;

      494       }

      495     }

      496   }

      497 #endif // !AMD64

      498 

      499   if (stub != NULL) {

      500     // save all thread context in case we need to restore it

      501     if (thread != NULL) thread->set_saved_exception_pc(pc);

      502 

      503     uc->uc_mcontext.gregs[REG_PC] = (greg_t)stub;

      504     return true;

      505   }

      506 

      507   // signal-chaining

      508   if (os::Linux::chained_handler(sig, info, ucVoid)) {

      509      return true;

      510   }

      511 

      512   if (!abort_if_unrecognized) {

      513     // caller wants another chance, so give it to him

      514     return false;

      515   }

      516 

      517   if (pc == NULL && uc != NULL) {

      518     pc = os::Linux::ucontext_get_pc(uc);

      519   }

      520 

      521   // unmask current signal

      522   sigset_t newset;

      523   sigemptyset(&newset);

      524   sigaddset(&newset, sig);

      525   sigprocmask(SIG_UNBLOCK, &newset, NULL);

      526 

      527   VMError err(t, sig, pc, info, ucVoid);

      528   err.report_and_die();

      529 

      530   ShouldNotReachHere();

      531 }

      532 

      533 void os::Linux::init_thread_fpu_state(void) {

      534 #ifndef AMD64

      535   // set fpu to 53 bit precision

      536   set_fpu_control_word(0x27f);

      537 #endif // !AMD64

      538 }

      539 

      540 int os::Linux::get_fpu_control_word(void) {

      541 #ifdef AMD64

      542   return 0;

      543 #else

      544   int fpu_control;

      545   _FPU_GETCW(fpu_control);

      546   return fpu_control & 0xffff;

      547 #endif // AMD64

      548 }

      549 

      550 void os::Linux::set_fpu_control_word(int fpu_control) {

      551 #ifndef AMD64

      552   _FPU_SETCW(fpu_control);

      553 #endif // !AMD64

      554 }

      555 

      556 // Check that the linux kernel version is 2.4 or higher since earlier

      557 // versions do not support SSE without patches.

      558 bool os::supports_sse() {

      559 #ifdef AMD64

      560   return true;

      561 #else

      562   struct utsname uts;

      563   if( uname(&uts) != 0 ) return false; // uname fails?

      564   char *minor_string;

      565   int major = strtol(uts.release,&minor_string,10);

      566   int minor = strtol(minor_string+1,NULL,10);

      567   bool result = (major > 2 || (major==2 && minor >= 4));

      568 #ifndef PRODUCT

      569   if (PrintMiscellaneous && Verbose) {

      570     tty->print("OS version is %d.%d, which %s support SSE/SSE2\n",

      571                major,minor, result ? "DOES" : "does NOT");

      572   }

      573 #endif

      574   return result;

      575 #endif // AMD64

      576 }

      577 

      578 bool os::is_allocatable(size_t bytes) {

      579 #ifdef AMD64

      580   // unused on amd64?

      581   return true;

      582 #else

      583 

      584   if (bytes < 2 * G) {

      585     return true;

      586   }

      587 

      588   char* addr = reserve_memory(bytes, NULL);

      589 

      590   if (addr != NULL) {

      591     release_memory(addr, bytes);

      592   }

      593 

      594   return addr != NULL;

      595 #endif // AMD64

      596 }

      597 

      598 ////////////////////////////////////////////////////////////////////////////////

      599 // thread stack

      600 

      601 #ifdef AMD64

      602 size_t os::Linux::min_stack_allowed  = 64 * K;

      603 

      604 // amd64: pthread on amd64 is always in floating stack mode

      605 bool os::Linux::supports_variable_stack_size() {  return true; }

      606 #else

      607 size_t os::Linux::min_stack_allowed  =  (48 DEBUG_ONLY(+4))*K;

      608 

      609 #ifdef __GNUC__

      610 #define GET_GS() ({int gs; __asm__ volatile("movw %%gs, %w0":"=q"(gs)); gs&0xffff;})

      611 #endif

      612 

      613 // Test if pthread library can support variable thread stack size. LinuxThreads

      614 // in fixed stack mode allocates 2M fixed slot for each thread. LinuxThreads

      615 // in floating stack mode and NPTL support variable stack size.

      616 bool os::Linux::supports_variable_stack_size() {

      617   if (os::Linux::is_NPTL()) {

      618      // NPTL, yes

      619      return true;

      620 

      621   } else {

      622     // Note: We can't control default stack size when creating a thread.

      623     // If we use non-default stack size (pthread_attr_setstacksize), both

      624     // floating stack and non-floating stack LinuxThreads will return the

      625     // same value. This makes it impossible to implement this function by

      626     // detecting thread stack size directly.

      627     //

      628     // An alternative approach is to check %gs. Fixed-stack LinuxThreads

      629     // do not use %gs, so its value is 0. Floating-stack LinuxThreads use

      630     // %gs (either as LDT selector or GDT selector, depending on kernel)

      631     // to access thread specific data.

      632     //

      633     // Note that %gs is a reserved glibc register since early 2001, so

      634     // applications are not allowed to change its value (Ulrich Drepper from

      635     // Redhat confirmed that all known offenders have been modified to use

      636     // either %fs or TSD). In the worst case scenario, when VM is embedded in

      637     // a native application that plays with %gs, we might see non-zero %gs

      638     // even LinuxThreads is running in fixed stack mode. As the result, we'll

      639     // return true and skip _thread_safety_check(), so we may not be able to

      640     // detect stack-heap collisions. But otherwise it's harmless.

      641     //

      642 #ifdef __GNUC__

      643     return (GET_GS() != 0);

      644 #else

      645     return false;

      646 #endif

      647   }

      648 }

      649 #endif // AMD64

      650 

      651 // return default stack size for thr_type

      652 size_t os::Linux::default_stack_size(os::ThreadType thr_type) {

      653   // default stack size (compiler thread needs larger stack)

      654 #ifdef AMD64

      655   size_t s = (thr_type == os::compiler_thread ? 4 * M : 1 * M);

      656 #else

      657   size_t s = (thr_type == os::compiler_thread ? 2 * M : 512 * K);

      658 #endif // AMD64

      659   return s;

      660 }

      661 

      662 size_t os::Linux::default_guard_size(os::ThreadType thr_type) {

      663   // Creating guard page is very expensive. Java thread has HotSpot

      664   // guard page, only enable glibc guard page for non-Java threads.

      665   return (thr_type == java_thread ? 0 : page_size());

      666 }

      667 

      668 // Java thread:

      669 //

      670 //   Low memory addresses

      671 //    +------------------------+

      672 //    |                        |\  JavaThread created by VM does not have glibc

      673 //    |    glibc guard page    | - guard, attached Java thread usually has

      674 //    |                        |/  1 page glibc guard.

      675 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()

      676 //    |                        |\

      677 //    |  HotSpot Guard Pages   | - red and yellow pages

      678 //    |                        |/

      679 //    +------------------------+ JavaThread::stack_yellow_zone_base()

      680 //    |                        |\

      681 //    |      Normal Stack      | -

      682 //    |                        |/

      683 // P2 +------------------------+ Thread::stack_base()

      684 //

      685 // Non-Java thread:

      686 //

      687 //   Low memory addresses

      688 //    +------------------------+

      689 //    |                        |\

      690 //    |  glibc guard page      | - usually 1 page

      691 //    |                        |/

      692 // P1 +------------------------+ Thread::stack_base() - Thread::stack_size()

      693 //    |                        |\

      694 //    |      Normal Stack      | -

      695 //    |                        |/

      696 // P2 +------------------------+ Thread::stack_base()

      697 //

      698 // ** P1 (aka bottom) and size ( P2 = P1 - size) are the address and stack size returned from

      699 //    pthread_attr_getstack()

      700 

      701 static void current_stack_region(address * bottom, size_t * size) {

      702   if (os::Linux::is_initial_thread()) {

      703      // initial thread needs special handling because pthread_getattr_np()

      704      // may return bogus value.

      705      *bottom = os::Linux::initial_thread_stack_bottom();

      706      *size   = os::Linux::initial_thread_stack_size();

      707   } else {

      708      pthread_attr_t attr;

      709 

      710      int rslt = pthread_getattr_np(pthread_self(), &attr);

      711 

      712      // JVM needs to know exact stack location, abort if it fails

      713      if (rslt != 0) {

      714        if (rslt == ENOMEM) {

      715          vm_exit_out_of_memory(0, "pthread_getattr_np");

      716        } else {

      717          fatal(err_msg("pthread_getattr_np failed with errno = %d", rslt));

      718        }

      719      }

      720 

      721      if (pthread_attr_getstack(&attr, (void **)bottom, size) != 0) {

      722          fatal("Can not locate current stack attributes!");

      723      }

      724 

      725      pthread_attr_destroy(&attr);

      726 

      727   }

      728   assert(os::current_stack_pointer() >= *bottom &&

      729          os::current_stack_pointer() < *bottom + *size, "just checking");

      730 }

      731 

      732 address os::current_stack_base() {

      733   address bottom;

      734   size_t size;

      735   current_stack_region(&bottom, &size);

      736   return (bottom + size);

      737 }

      738 

      739 size_t os::current_stack_size() {

      740   // stack size includes normal stack and HotSpot guard pages

      741   address bottom;

      742   size_t size;

      743   current_stack_region(&bottom, &size);

      744   return size;

      745 }

      746 

      747 /////////////////////////////////////////////////////////////////////////////

      748 // helper functions for fatal error handler

      749 

      750 void os::print_context(outputStream *st, void *context) {

      751   if (context == NULL) return;

      752 

      753   ucontext_t *uc = (ucontext_t*)context;

      754   st->print_cr("Registers:");

      755 #ifdef AMD64

      756   st->print(  "RAX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RAX]);

      757   st->print(", RBX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RBX]);

      758   st->print(", RCX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RCX]);

      759   st->print(", RDX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RDX]);

      760   st->cr();

      761   st->print(  "RSP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RSP]);

      762   st->print(", RBP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RBP]);

      763   st->print(", RSI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RSI]);

      764   st->print(", RDI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RDI]);

      765   st->cr();

      766   st->print(  "R8 =" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R8]);

      767   st->print(", R9 =" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R9]);

      768   st->print(", R10=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R10]);

      769   st->print(", R11=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R11]);

      770   st->cr();

      771   st->print(  "R12=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R12]);

      772   st->print(", R13=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R13]);

      773   st->print(", R14=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R14]);

      774   st->print(", R15=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_R15]);

      775   st->cr();

      776   st->print(  "RIP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_RIP]);

      777   st->print(", EFLAGS=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EFL]);

      778   st->print(", CSGSFS=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_CSGSFS]);

      779   st->print(", ERR=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_ERR]);

      780   st->cr();

      781   st->print("  TRAPNO=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_TRAPNO]);

      782 #else

      783   st->print(  "EAX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EAX]);

      784   st->print(", EBX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EBX]);

      785   st->print(", ECX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_ECX]);

      786   st->print(", EDX=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EDX]);

      787   st->cr();

      788   st->print(  "ESP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_UESP]);

      789   st->print(", EBP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EBP]);

      790   st->print(", ESI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_ESI]);

      791   st->print(", EDI=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EDI]);

      792   st->cr();

      793   st->print(  "EIP=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EIP]);

      794   st->print(", EFLAGS=" INTPTR_FORMAT, uc->uc_mcontext.gregs[REG_EFL]);

      795   st->print(", CR2=" INTPTR_FORMAT, uc->uc_mcontext.cr2);

      796 #endif // AMD64

      797   st->cr();

      798   st->cr();

      799 

      800   intptr_t *sp = (intptr_t *)os::Linux::ucontext_get_sp(uc);

      801   st->print_cr("Top of Stack: (sp=" PTR_FORMAT ")", sp);

      802   print_hex_dump(st, (address)sp, (address)(sp + 8*sizeof(intptr_t)), sizeof(intptr_t));

      803   st->cr();

      804 

      805   // Note: it may be unsafe to inspect memory near pc. For example, pc may

      806   // point to garbage if entry point in an nmethod is corrupted. Leave

      807   // this at the end, and hope for the best.

      808   address pc = os::Linux::ucontext_get_pc(uc);

      809   st->print_cr("Instructions: (pc=" PTR_FORMAT ")", pc);

      810   print_hex_dump(st, pc - 32, pc + 32, sizeof(char));

      811 }

      812 

      813 void os::print_register_info(outputStream *st, void *context) {

      814   if (context == NULL) return;

      815 

      816   ucontext_t *uc = (ucontext_t*)context;

      817 

      818   st->print_cr("Register to memory mapping:");

      819   st->cr();

      820 

      821   // this is horrendously verbose but the layout of the registers in the

      822   // context does not match how we defined our abstract Register set, so

      823   // we can't just iterate through the gregs area

      824 

      825   // this is only for the "general purpose" registers

      826 

      827 #ifdef AMD64

      828   st->print("RAX="); print_location(st, uc->uc_mcontext.gregs[REG_RAX]);

      829   st->print("RBX="); print_location(st, uc->uc_mcontext.gregs[REG_RBX]);

      830   st->print("RCX="); print_location(st, uc->uc_mcontext.gregs[REG_RCX]);

      831   st->print("RDX="); print_location(st, uc->uc_mcontext.gregs[REG_RDX]);

      832   st->print("RSP="); print_location(st, uc->uc_mcontext.gregs[REG_RSP]);

      833   st->print("RBP="); print_location(st, uc->uc_mcontext.gregs[REG_RBP]);

      834   st->print("RSI="); print_location(st, uc->uc_mcontext.gregs[REG_RSI]);

      835   st->print("RDI="); print_location(st, uc->uc_mcontext.gregs[REG_RDI]);

      836   st->print("R8 ="); print_location(st, uc->uc_mcontext.gregs[REG_R8]);

      837   st->print("R9 ="); print_location(st, uc->uc_mcontext.gregs[REG_R9]);

      838   st->print("R10="); print_location(st, uc->uc_mcontext.gregs[REG_R10]);

      839   st->print("R11="); print_location(st, uc->uc_mcontext.gregs[REG_R11]);

      840   st->print("R12="); print_location(st, uc->uc_mcontext.gregs[REG_R12]);

      841   st->print("R13="); print_location(st, uc->uc_mcontext.gregs[REG_R13]);

      842   st->print("R14="); print_location(st, uc->uc_mcontext.gregs[REG_R14]);

      843   st->print("R15="); print_location(st, uc->uc_mcontext.gregs[REG_R15]);

      844 #else

      845   st->print("EAX="); print_location(st, uc->uc_mcontext.gregs[REG_EAX]);

      846   st->print("EBX="); print_location(st, uc->uc_mcontext.gregs[REG_EBX]);

      847   st->print("ECX="); print_location(st, uc->uc_mcontext.gregs[REG_ECX]);

      848   st->print("EDX="); print_location(st, uc->uc_mcontext.gregs[REG_EDX]);

      849   st->print("ESP="); print_location(st, uc->uc_mcontext.gregs[REG_ESP]);

      850   st->print("EBP="); print_location(st, uc->uc_mcontext.gregs[REG_EBP]);

      851   st->print("ESI="); print_location(st, uc->uc_mcontext.gregs[REG_ESI]);

      852   st->print("EDI="); print_location(st, uc->uc_mcontext.gregs[REG_EDI]);

      853 #endif // AMD64

      854 

      855   st->cr();

      856 }

      857 

      858 void os::setup_fpu() {

      859 #ifndef AMD64

      860   address fpu_cntrl = StubRoutines::addr_fpu_cntrl_wrd_std();

      861   __asm__ volatile (  "fldcw (%0)" :

      862                       : "r" (fpu_cntrl) : "memory");

      863 #endif // !AMD64

      864 }